Machine Learning-Based Physical Activity Tracking with a view to Frailty Analysis.

Frailty in old age is defined as the individual intrinsic susceptibility of having bad outcomes following a health problem. It relies on sarcopenia, mobility and activity. Recognizing and monitoring a range of physical activities is a necessary step which precedes the analysis of this syndrome. This paper investigates the optimal tools for this recognition in terms of type and placement of wearable sensors. Two machine learning procedures are proposed and compared on a public dataset. The first one is based on deep learning, where feature extraction is done manually, by constructing activity images from raw signals and applying convolutional neural networks to learn optimal features from these images. The second one is based on shallow learning, where hundreds of handcrafted features are extracted manually, followed by a novel feature selection approach to retain the most discriminant subset.Clinical relevance- This analysis is an indispensable prerequisite to develop efficacious way in order to identify people with frailty using sensors and moreover, to take on the challenge of frailty prevention, an actual world health organization priority.

Linear and nonlinear causality between signals: methods, examples and neurophysiological applications.

In this paper, we will present and review the most usual methods to detect linear and nonlinear causality between signals: linear Granger causality test (Geweke in J Am Stat Assoc 77:304-313, 1982) extended to direct causality in multivariate case (LGC), directed coherence (DCOH, Saito and Harashima in Recent advances in EEG and EMG data processing, Elsevier, Amsterdam, 1981), partial directed coherence (PDC, Sameshima and Baccala 1999) and nonlinear Granger causality test of Baek and Brock (in Working Paper University of Iowa, 1992) extended to direct causality in multivariate case (partial nonlinear Granger causality, PNGC). All these methods are tested and compared on several ARX, Poisson and nonlinear models, and on neurophysiological data (depth EEG). The results show that LGC, DCOH and PDC are not very robust in relation to nonlinear linkages but they seem to correctly find linear linkages if only the autoregressive parts are nonlinear. PNGC is extremely dependent on the choice of parameters. Moreover, LGC and PNGC may give misleading results in the case of causality on a spectral band, which is illustrated by our neurophysiological database.